PROJECT SUMMARY Reward-guided decision-making and impulse control are disrupted after chronic cocaine use. These changes have been attributed to altered function in brain circuits critical for computations of reward predictions and action policies. `Reward prediction' signals reflect the reward the animal expects to receive as a result of behavior, thus reflecting goals associated with decisions. `Action policies' are rules that govern behavior that at triggered by external stimuli or context, and are thought to underlie habits. Both reward predictions and action policies are modified when there are violations in predictions known as `reward prediction errors'. `Signed' reward prediction errors reflect the valence associated with an error, strengthening or weakening the associability between cues and outcomes/responses. `Unsigned' prediction errors reflect the surprise induced by errors which lead to increases in attention so that learning can occur. We have uncovered neural correlates of these constructs and the relationship between them by recording from multiple brain areas as rats perform an odor guided decision-making task in which we unexpectedly varied the delay to and size of reward across several trial blocks. We have shown that nucleus accumbens core (NAc) encodes reward predictions, firing strongly for cues that predict more valued reward, whereas firing in dorsal lateral striatum (DLS) is highly associative, encoding action policies such as stimulus-response associations and contextual bias signals (e.g., in this context bias choices to the right). We have also shown that midbrain dopamine (DA) neurons increase firing to unexpected reward and decrease firing to unexpected reward omission. During learning these signed prediction errors transfer to cues, with cues predicting more valued reward eliciting stronger firing. Unlike firing of DA neurons, our work has shown that firing in anterior cingulate cortex (ACC) better reflects an integrated unsigned reward prediction error signal, increasing during unexpected up- and down-shifts in value at the time of the error and during cue sampling on subsequent trials. This work suggests a model by which DA reward prediction errors modify reward prediction signals in NAc and action policy signals in DLS, while ACC increases attention toward stimuli after violations in reward prediction (signaled by DA) so that learning can occur. Cocaine exposure impairs reward prediction signals and prediction error signals in NAc and DA neurons, while increasing the prevalence of contextual action policies in DLS. In Aim 1 we propose to restore the cocaine induced imbalance of processing between NAc and DLS by repairing DA signals via optogenetics. In Aim 2 we will determine if attention and error correlates in ACC are altered after cocaine exposure. Finally, in Aim 3, we will determine how ACC and DA neurons interact during the computation of errors and the development of cue selectivity. By performing these experiments we will gain further insight into how the brain functions normally, how it is disrupted after chronic cocaine use, and determine if repairing neural signals might restore behavior and neural constructs in downstream regions to normal levels.